The thesis deals with finding the worst subgroups in the forecasts of a machine learning model for fruits and vegetables. The primary goal is the improvement of the model, by seeing where it made a mistake, then analyzing that mistake and attempting to learn why it happened. We solved the problem by defining a process that searches for critical subgroups, first gathering and preparing the data, then running an algorithm to find a few problematic subgroups. Beside that, another part of problem solving was analyzing the cases themselves, to further improve the process. After the implementation, the process runs weekly and is used for the business needs of the company.
|
